![]()
Secure Code Warrior, a leader in AI software governance and secure coding training, today released the second wave of findings from the SCW AI Trust Index, a living benchmark that measures how often leading AI models generate code with security vulnerabilities. After evaluating more than 30 models from OpenAI, Anthropic, Google, Alibaba, Mistral and others, the research found that AI models have improved in the last 12 months, write markedly less vulnerable code when given curated security rules, and that a budget model can carry less risk than a flagship. Secure Code Warrior updated its methodology to test models within a harness, giving a more accurate picture of real-world use and showing how security instructions and harness prompts affect performance. This refreshed methodology also includes data comparisons on the functional completeness of AI-generated code, offering a more complete picture of overall risk and efficiency for software development tasks across models.
In total, the company used AI to build more than 15,000 apps and 41.3 million lines of code in 11 coding languages throughout the course of this research.
Frontier LLMs are writing a growing share of developers’ code. Sonar’s 2026 State of Code Developer Survey finds that AI is expected to account for 65% of committed code by 2027. Meanwhile, new models arrive so quickly that the rankings can change within weeks. In Secure Code Warrior’s AI Trust Index, the AI models tested produced between 0.24 and 3.66 vulnerabilities per 1,000 lines of code, a 15-fold difference between the best- and worst-performing models. While methodologies differ, separate research from Stanford University found that human-only commits introduced roughly 0.08 vulnerabilities per 1,000 lines.
Security leaders need a clear view of the risks their teams are inheriting, and whether those risks shift depending on the models and frameworks being leveraged. The SCW AI Trust Index provides that view through empirical measurement that evolves alongside new models, keeping pace with the latest developments and tools available to developer teams. Secure Code Warrior continually tests new models, updating the results promptly following each release. The latest version of the Trust Index is meant to help organizations evaluate how the models their employees rely on measure up and understand what improves them. The research shows that governance of the AI coding stack is where security is won. Key findings include:
- Vague instruction does little, while curated rules cut measured risk by up to 70%: The default model sets the starting point; the prompt decides how secure AI-generated code ends up.
- No model is strongest everywhere. 15 different models rank first, or tie for first, in at least one of the 11 frameworks at baseline. GPT-6-Astra-High leads or ties in the most: 4.
- Rulesets help the stronger models more. On the lower-risk half of models, both SCW rules and third-party rules cut risk by an average of 54.15%; on the higher-risk half, an average of 34.05%.
- A budget model can beat a flagship on security: Some budget models, with or without the right rules, carry less measured risk than premium models on their own. For example, GPT 6 Luna carries a mean risk of 19.4 at $0.10 a run, while Claude Opus 5.5 carries a mean risk of 43.6 at $8.17 a run (models alone, no rules).
“It’s easy to assume every new AI model writes safer code than the last, or a frontier model is more secure than a budget model. Our updated findings show this isn’t always the case,” said Pieter Danhieux, Secure Code Warrior Co-Founder & CEO. “Today’s models know approaches to secure coding, but they don’t do it by default. The harness, prompts and rules they’re given decide how secure the output is, and that behavior can shift with every new release. A one-time benchmark can’t keep pace, which is why the SCW AI Trust Index evolves alongside the models themselves. Security leaders and developers need a current, evidence-based view of the risks they’re inheriting that is verifiable by the languages and frameworks in use within the development team.”
AI code generation is here to stay. Developers and IT teams face increasing pressure to bring new AI-enabled apps to market, but to do so safely, they need full visibility into the models they rely on. Every model has a distinct fingerprint. The SCW AI Trust Index helps developers benchmark model behavior and understand how to best leverage each model’s capabilities, identify where risk is introduced, and produce more secure code from the outset.
To access the SCW AI Trust Index, please visit: https://www.securecodewarrior.com/product/ai-trust-index.
About Secure Code Warrior
Secure Code Warrior is a leader in AI software governance and developer security upskilling, enabling enterprises to control AI-driven software development across the SDLC. Built on a decade of developer security expertise, it delivers AI visibility, policy enforcement, and targeted learning to prevent vulnerabilities and strengthen software quality before production.
View source version on businesswire.com: https://www.businesswire.com/news/home/20261117306637/en/
Media gallery

